Everyone told you to add schema markup, so AI tools would cite your site. A clever experiment with fake duck-company data suggests they might be wrong.
Also Read: Why Reddit Is So Powerful for SaaS Keyword Research
Topic: Schema Markup, LLMs, SaaS SEO·
Reading time: 7 min·Based on research by Mark Williams-Cook, Charles Floate & John Mueller
SaasPedia TL;DR Box
Schema markup is not being used by LLMs the way it was designed.
AI tools like ChatGPT just read your HTML like regular text — valid schema or not.
For SaaS sites, your real focus should be on
well-written body content, not background code.
If you run a SaaS company, you’ve probably heard this advice: “Add schema markup to your website — it helps AI tools like ChatGPT and Perplexity understand your content and cite you more often.”
It sounds logical. A whole movement called Generative Engine Optimization (GEO) has been built around this idea. But a simple, brilliant experiment may have just pulled the rug out from under all of it.
First: What Is Schema Markup?
Schema markup is code you add behind the scenes on your website. Visitors never see it — but search engines and AI tools are supposed to read it. It labels your content so machines can understand it more easily.
For example, instead of just writing “Call us at 555-1234” in plain text, you’d also add that phone number in a special coded format. Google could then say, “Got it — that’s a phone number for a local business.”
The idea was smart when it was invented. But the web has changed a lot since then.
The Duck Test Experiment
SEO expert Mark Williams-Cook decided to put the schema-for-LLMs theory to the test. His approach? Make up an entirely fictional company with a completely nonsensical schema — and see what AI does with it.
The Experiment – SaasPedia
Case Study
The Experiment
Meet Duckier T-Shirts — a company that doesn’t exist
Mark built a webpage for a fake company and filled it with schema he completely invented.
None of it followed any real schema rules.
Here’s what the fake schema looked like:
{
“@type”: “MallardEnterprise”,
“flockName”: “Duckia T-Shirts”,
“waddleStyle”: “Aggressive”,
“nestingGrounds”: “77 The Muddy Bank”,
“region”: “South Pawnure”,
“featherCode”: “DK99”,
“country”: “United Queendom”,
“quackVolume”: “Loud”
}
None of these are real schema types. “Waddle style” and “quack volume” are not things.
Mark made them up entirely.
Then Mark asked ChatGPT and Perplexity the same question: “What is the address of this company?” — and gave them the URL of the fake page.
AI Response – SaasPedia
AI Response (ChatGPT & Perplexity)
What the AI Extracted from the Fake Schema
“The address shown in this page’s structured data is:
77 The Muddy Bank, South Pawnure, DK99, United Queendom.
”
Even though the schema properties were completely invented,
the AI still extracted and formatted the address as if it were valid.
Both AI tools confidently read the fake, made-up, totally invalid schema… and treated it like a real address.
What This Actually Means
SaasPedia Quote Block
“
Schema is not being used in the explicit sense it was designed for within LLMs.
— Mark Williams-Cook
The conclusion is clear: LLMs don’t parse schema as structured data. They just scan everything in your HTML — code, tags, text, schema — and pull out whatever looks relevant to the question being asked.
It doesn’t matter if your schema is valid. It doesn’t matter if you follow the rules. The AI isn’t reading your schema the way a developer reads code. It’s reading it the same way it reads a paragraph in your blog post.
SaaSPedia Key Insights
Key Takeaways from the Experiment
✓
LLMs treat schema markup like regular HTML text — nothing special happens.
✓
Invalid, made-up schema gets picked up just as easily as valid schema.
✓
Adding schema won’t give you an advantage in AI citations over competitors.
✓
Body content — your actual paragraphs — is what AI tools prioritize most.
Why Schema Was Created (And Why LLMs Don’t Need It)
To understand why this matters, we have to understand the original problem schema was solving.
Back when the schema was created, search engine algorithms were pretty basic. They could read words, but they struggled to understand context. Was “Mercury” a planet, a car brand, or a Greek god? Was “1600 Pennsylvania Avenue” an address or a reference in a news article?
Schema was the solution. You’d label things so the algorithm could categorize them without guessing.
But modern large language models don’t have that problem. They’re trained on vast amounts of text and can already understand context, meaning, and relationships in natural language. They don’t need labels — they can figure it out on their own.
With an LLM, it doesn’t need to interpret defined variables because it can already interpret the data from the
exact content. They seem to prioritize
body text and body content over every other part of the web page.
Charles Floate
SEO Strategist
Does Schema Still Matter for Google SEO?
Yes — but only in specific, limited ways. Google’s own public-facing search expert, John Mueller, was direct about it:
Structured data won’t make your site rank better.
It is used for displaying search features.
John Mueller
Google Search Advocate
Schema can help you get rich results — the visual extras that show up in Google search like star ratings, FAQ dropdowns, recipe cards, or product pricing. These can improve how often people click your result, which indirectly helps your rankings.
But schema alone doesn’t boost rankings. And Google ignores schema frequently — especially if no competitors in your search results are using it to get those rich features.
| Schema Use Case |
Does It Help? |
Notes |
| Getting cited by LLMs (ChatGPT, Perplexity) |
No |
Treated as regular text |
| Google rankings (direct boost) |
No |
Confirmed by John Mueller |
| Google rich results (visual features) |
Sometimes |
Only if competitors already use it |
| Entity trust for new sites (SameAs) |
Yes |
Helps Google recognize your brand |
The One Schema That’s Still Worth Using: SameAs
There’s one specific type of schema that SEO experts still recommend — especially for new SaaS sites. It’s called SameAs schema.
SameAs schema tells Google: “This website is the same company as this LinkedIn page, this Twitter account, this Crunchbase listing…” It connects your website to your external profiles and social accounts.
This doesn’t directly help you get cited by AI. But it does help Google recognize and trust your brand faster — which matters a lot when you’re a new company trying to build credibility.
SEO strategist Charles Floate recommends linking your most important profiles: Twitter/X, LinkedIn, YouTube, Facebook, and an About.me page. Keep it to 5–10 profiles max — stuffing in 50+ links will likely get ignored.
What SaaS Sites Should Actually Do
If schema isn’t the answer for AI visibility, what is? The research points to one clear winner: your body content.
Here’s exactly what to focus on:
- Write clear, specific body contentPut your key information in plain text where both humans and AI can actually read it. Don’t hide important facts in background code. If you want AI to say “This tool saves 3 hours per week,” then write that exact sentence on your page.
- Be concrete about what your product doesVague copy like “We help businesses grow” won’t get picked up by AI. Specific statements like “Our invoicing software automates billing for freelancers and reduces payment delays by 40%” will.
- Build external credibilityAI tools don’t just read your website. They pull from reviews, press coverage, blog mentions, and third-party sites. Getting featured on industry blogs, product review platforms, and in press articles massively increases your citation chances.
- Check what competitors are already doingLook at Google search results for your target keywords. If competitors are getting rich result features from schema (like FAQ dropdowns or star ratings), match their schema to compete. If nobody’s using it, don’t bother for AI purposes.
- Add SameAs schema if you’re a new siteConnect your website to your social profiles and key external pages. This builds Google’s trust in your entity faster — which can unlock rankings in competitive niches more quickly.
⚠️
Warning: Don’t Try to Hide Text for AI
You might think: “If AI reads all the HTML, I could hide text in the code that humans don’t see but AI picks up.”
Technically possible — but Google treats hidden text as spam and will penalize your site.
Not worth the risk.
The Bottom Line
The Duck Test experiment made one thing very clear: LLMs treat schema like any other text on your page. It doesn’t matter if your schema is valid, perfectly structured, or follows every rule. The AI is simply reading your HTML and pulling out whatever looks relevant.
For SaaS sites, this means the path to getting cited by AI tools is the same as the path to good SEO: write clear, specific, genuinely useful content. Put your most important information in your actual paragraphs — not hidden in background code.
Schema for Google rich results? Still situationally useful. SameAs schema for new sites? Still a smart move for building entity trust. But as a magic ticket to AI citations? The duck has left the building.
Sources & Credits
This article is based on research and commentary from
SEO expert Mark Williams-Cook (Duck Test experiment),
Charles Floate (entity schema strategy), and
Google Search Advocate John Mueller (structured data statement).
Referenced from Episode 956 of The Edward Show podcast.
